Thermodynamic factors determining the oxygen vacancy formation energies of oxide perovskites

Efficient and accurate methods for predicting oxygen vacancy formation (OVF) energies could significantly advance materials for solar thermochemical H₂ production and solid oxide fuel cells. While current, mostly linear, models based on O 2p band centers and formation enthalpies are efficient enough for high-throughput screening, they are accurate only for a small subset of materials. Here, we introduce a linear model, based entirely on local thermodynamic features, that is efficient and accurate for a diverse set of ABO₃ perovskites containing insulators and metals, six lattice systems, five A-site cations including the redox-active Ce, and the seven 3d transition metals from Ti to Co. First, we constructed a database of neutral OVF energies in these perovskites using state-of-the-art density functional theory with the strongly constrained and appropriately normed exchange-correlation functional and Hubbard U corrections. Then, we generated models, including as a feature our newly developed crystal bond dissociation energies, using genetic programming and regularized linear regression with mean absolute errors lower than 0.50 eV for stable systems. Unlike current methods in the literature, our best model provides an imminently intuitive physical picture for OVF.